[lsan] Relax check for allocator_end in ProcessGlobalRegionsCallback.

[m.ostapenko]

Summary

While testing I hit on CHECK_LE(allocator_end, end) randomly from one build to another. Looking to GetAllocatorGlobalRange(&allocator_begin, &allocator_end) just few lines above this check, It appears that allocator_end theoretically may become equal to end (if allocator locates in top addresses of corresponding global section).
Just relax this check to avoid such failure.

[kcc]

LGTM

compiler-rt/trunk/lib/lsan/lsan_common_linux.cc

if (!(phdr->p_flags & PF_W) || (phdr->p_type != PT_LOAD) ||
         (phdr->p_memsz == 0))
       continue;
     uptr begin = info->dlpi_addr + phdr->p_vaddr;
     uptr end = begin + phdr->p_memsz;
     uptr allocator_begin = 0, allocator_end = 0;
     GetAllocatorGlobalRange(&allocator_begin, &allocator_end);
     if (begin <= allocator_begin && allocator_begin < end) {
       CHECK_LE(allocator_begin, allocator_end);
-      CHECK_LT(allocator_end, end);
+      CHECK_LE(allocator_end, end);
       if (begin < allocator_begin)
         ScanRangeForPointers(begin, allocator_begin, frontier, "GLOBAL",
                              kReachable);
       if (allocator_end < end)
         ScanRangeForPointers(allocator_end, end, frontier, "GLOBAL",
                              kReachable);
     } else {
       ScanRangeForPointers(begin, end, frontier, "GLOBAL", kReachable);